A Genotype-Phenotype Correlation Study of SHV β-Lactamases Offers New Insight into SHV Resistance Profiles

The SHV β-lactamases (BLs) have undergone strong allele diversification that has changed their substrate specificities. Based on 147 NCBI entries for SHV alleles, in silico mathematical models predicted 5 positions as relevant for the β-lactamase inhibitor (BLI)-resistant (2br) phenotype, 12 positions as relevant for the extended-spectrum BL (ESBL) (2be) phenotype, and 2 positions as related solely to the narrow-spectrum (2b) phenotype. These positions and six additional positions described in other studies (including one promoter mutation) were systematically substituted and investigated for their substrate specificities in a BL-free Escherichia coli background, representing, to our knowledge, the most comprehensive substrate and substitution analysis for SHV alleles to date.

ABSTRACT

The SHV β-lactamases (BLs) have undergone strong allele diversification that has changed their substrate specificities. Based on 147 NCBI entries for SHV alleles, in silico mathematical models predicted 5 positions as relevant for the β-lactamase inhibitor (BLI)-resistant (2br) phenotype, 12 positions as relevant for the extended-spectrum BL (ESBL) (2be) phenotype, and 2 positions as related solely to the narrow-spectrum (2b) phenotype. These positions and six additional positions described in other studies (including one promoter mutation) were systematically substituted and investigated for their substrate specificities in a BL-free Escherichia coli background, representing, to our knowledge, the most comprehensive substrate and substitution analysis for SHV alleles to date. An in vitro analysis confirmed the essentiality of positions 238 and 179 for the 2be phenotype and of position 69 for the 2br phenotype. The E240K and E240R substitutions, which do not occur alone in known 2br SHV variants, led to a 2br phenotype, indicating a latent BLI resistance potential of these substitutions. The M129V, A234G, S271I, and R292Q substitutions conferred latent resistance to cefotaxime. In addition, seven positions that were found not always to be associated with the ESBL phenotype resulted in increased resistance to ceftaroline. We also observed that coupling of a strong promoter (IS26) to an A146V mutant with the 2b phenotype resulted in highly increased resistance to BLIs, cefepime, and ceftaroline but not to third-generation cephalosporins, indicating that SHV enzymes represent an underestimated risk for empirical therapies that use piperacillin-tazobactam or cefepime to treat different infectious diseases caused by Gram-negative bacteria.

INTRODUCTION

The sulfhydryl-variable (SHV) β-lactamases (BLs) belong to the phylogenetic Ambler class A BLs (encoded by the bla_SHV gene) but can be subdivided according to their substrate and β-lactamase inhibitor (BLI) spectra based on the Bush-Jacoby-Medeiros (BJM) classification into the 2b type, which hydrolyzes penicillins and first- and second-generation cephalosporins (e.g., cefazolin [CFZ] or cefoxitin [FOX]); the 2be type, comprising extended-spectrum BLs (ESBLs) that hydrolyze third-generation cephalosporins (e.g., cefotaxime [CTX] or ceftazidime [CAZ]); and variants of the 2br type that are resistant to BLIs (e.g., clavulanic acid [CLA] or tazobactam [TAZ]).

The first plasmid-encoded SHV-1 BL in Escherichia coli, which most likely originated from a chromosomal homolog of Klebsiella pneumoniae, was identified in the 1970s (1), and the corresponding conjugative plasmid p453 was sequenced in 1988 (2). SHV-1 BLs exhibited high activity against ampicillin (AMP) (a third-generation penicillin) and first-generation cephalosporins, e.g., cephaloridine or cephalexin (3), but were inefficient against second-generation penicillins (oxacillin or cloxacillin). The first SHV-2 BL, located on the pBP60 plasmid, was isolated from a clinical sample in Germany in 1983 and showed only a few nucleotide mismatches relative to the bla_SHV-1 allele (4) but exhibited elevated activity against third-generation cephalosporins, e.g., CTX, CAZ, or ceftriaxone (CRO) (all with MICs of 4 mg/liter for SHV-2 versus <0.124 mg/liter for SHV-1 BLs) and aztreonam (ATM) (MIC_ATM, 1 mg/liter for SHV-2 versus 0.03 mg/liter for SHV-1).

The SHV BLs underwent strong allele diversification, spreading primarily in clinical Enterobacteriaceae isolates, with the highest prevalence in K. pneumoniae and E. coli, but also in other pathogens, such as Pseudomonas aeruginosa and Acinetobacter baumannii (5, 6). Of the 292 amino acid residues (Ambler numbering system [7]), 102 positions within the mature protein and 12 positions within the signal peptide (SP) exhibit variable residues; 8 variants show insertions within the sequence that were not considered in the 124 substituted positions.

The estimated number of uncharacterized SHV allelic variants with new amino acid substitutions, deletions, or insertions is increasing, but the molecular diagnosis of SHV variants is neglected (8) because only point mutations differentiate between the phenotypes, making them difficult to resolve by common PCR-based molecular assays. Microbiological testing is challenging because the ESBLs that are currently most prevalent belong to the CTX-M group (class A BLs that are not related to SHV alleles) and are often accompanied by TEM (related to SHV) or SHV BLs; thus, the phenotypes are overlapping. The presence of SHV alleles has achieved low clinical relevance, except for their BLI-resistant properties, which are assessed by antimicrobial susceptibility testing (AST). To develop a molecular differentiation assay based on single nucleotide triplets, it is necessary to know the phenotype-relevant substitutions. The impact of amino acid substitutions on the substrate spectrum of the SHV enzymes has been partially investigated for certain substrates and substitutions (9) or for selected variants (10) (for a review, see reference 11). Therefore, we applied different mathematical optimization algorithms and trained them by the known BJM-classified SHV variants to identify meaningful point mutations or mutation combinations within the SHV gene that distinguish between the 2b, 2be, and 2br phenotypes. To validate the predicted phenotype-relevant amino acid substitutions (PRASs), we performed site-directed mutagenesis of the SHV-1 gene, which was ligated into a BL-free cloning vector. The phenotypic resistance patterns of the mutated SHV derivatives were investigated in a BL-free E. coli background. The aims of this study were (i) to identify meaningful positions of the SHV gene that can be used in molecular diagnostic approaches to clearly differentiate between phenotypes and (ii) to determine whether such positions can be reliably predicted based on the available databases.

RESULTS

Phenotype-relevant amino acid positions and substitutions identified by mathematical models.

Among 146 SHV variants analyzed (with SHV-1 as the reference), 109 Ambler positions were subject to permutation. Two of the 146 variants (SHV-187 and SHV-188) possessed 2-amino-acid N-terminal elongations of the signal peptide; three additional variants contained insertions (SHV-100, SHV-188, and SHV-183). A total of 130 substitutions were present in the SHV BLs: 62 substitutions in known phenotypes and 32 in 2b, 39 in 2be, and 7 in 2br variants. The mean frequency of mutations (expressed as the sum of the mutations per group divided by the number of variants) was 2.3 for all variants (339/146) and the known phenotypes (168/72), 2.7 for 2b (54/28), 2.7 for 2be (105/39), and 1.8 for 2br (9/5).

By applying the model optimization approach optiSLang, 18 PRASs were predicted, and their relevance (expressed as coefficient of prognosis [CoP] values) to the specific BJM phenotypes was weighed. The modeling was performed for the SHV alleles, including those for the SP and using only the amino acid position matrix, or including the isoelectric points (pIs) of the corresponding amino acid residues and varying the analysis parameters (for details, see Materials and Methods). The results are provided in Fig. 1, and the averaged CoP values are summarized in Table 1. The best-fitting results were obtained by the Metamodel of Optimal Prognosis (MOP,) which automatically chooses the methods during the optimization process as well as using the manually set kriging regression model when pI is included. For the 2b and 2be phenotypes, the “leave-one-out” setting (for details, see Materials and Methods) achieved the best fit, while for the 2br phenotype, the automated validation setting was superior (for details, see Table S3 in the supplemental material). The best-fitting models sufficiently described only 41.1% of the 2b variants but as many as 50.9% of the 2be and 59.8% of the 2br variants. The majority of the positions were assigned as relevant for the 2be (n = 14) and 2b (n = 10) phenotypes, whereas only 4 positions were associated with the 2br phenotype (Fig. 1; Table S3).

FIG 1.

CoP values determined by optiSLang for the PRAS positions (according to the Ambler nomenclature) in SHV variants for the specific phenotypes 2b, 2be, and 2br. MOP, Metamodel of Optimal Prognosis; K, kriging; LOO, leave one out; –Z, excluding missing amino acid residues (set at zero); IP, including isoelectric points of the amino acid residues.

TABLE 1.

Averaged CoP values and corresponding scattering^a obtained for the PRASs by mathematical modeling using optiSLang

Amino acid position in SHV-1b	Amino acid residue in SHV-1	CoP value (± standard deviation) related to the following specific phenotype:			Amino acid(s) substituted	Drug(s) with significantly increased MIC(s)
		2b	2be	2br
8	I	0.015 (±0.002)		0.015 (±0.014)	F
25c	A				T, S	CPT
35	L	0.084 (±0.025)			Q	CPT
43c	R				S
69	M			0.135 (±0.017)	I, L	TAZ, CFPd
104	D		0.025 (±0.0)		G	CFP
112	H		0.022 (/)		Y
129c	M				V
142	V		0.025 (±0.001)		F
146c	A				T, V	CPT, CFPe
148	L	0.016 (±0.002)	0.026 (±0.002)		V	CPT
156c	G				D	CPT
169	L	0.03 (±0.001)	0.043 (±0.001)		Failed
179	D	0.029 (±0.0)	0.048 (±0.005)		N	CTX, CAZ, CRO, FEP
187	A		0.023 (/)		T
202	R		0.025 (±0.001)		S
234	K	0.031 (±0.003)	0.037 (/)	0.20 (±0.036)	R	AVI
235	T	0.016 (±0.001)		0.148 (±0.018)	A
238	G	0.275 (±0.042)	0.346 (±0.14)		S, A	CTX, CRO, CPD, FEP, CPT, ATM
240	E	0.161 (±0.034)	0.144 (±0.001)		K, R	CPT, TAZ
243	A	0.016 (±0.002)	0.025 (±0.001)		G	CFP
271	S		0.025 (±0.002)		I
292	R		0.05 (±0.002)		Q

^aThe scattering considers only models with CoP values and thus tends to zero if only a few models identified the corresponding position as relevant with similar CoP values; there is no scattering if only one model identified a selected residue as a PRAS (for details, see Fig. 1). A slash in parentheses indicates only one CoP value in one of the models.

^bListed according to the Ambler scheme.

^cAdditional mutants are based on the occurrence of the mutations noted in specific phenotypes.

^dThe MIC_CFP was >2-fold increased for the M69L variant.

^eThe MIC_CFP was >2-fold increased for the V146T mutant.

In general, only five positions showed averaged CoP values of >0.1, which can be interpreted as >10% relevance to the respective phenotype (Table 1). Positions 35 (CoP = 0.084) and 69 (CoP = 0.135) were assigned as exclusively relevant for the 2b and 2br phenotypes, respectively. Further strongly 2br-related positions were positions 234 (CoP = 0.20) and 235 (CoP = 0.148), while position 8 showed a CoP of only 0.015 for the 2br phenotype. Only two positions, 238 (CoP = 0.346) and 240 (CoP = 0.144), were assigned as strongly relevant for the 2be phenotype, but they seemed to be crucial for the 2b phenotype as well (CoP, 0.275 for position 238 and 0.161 for position 240).

The pBT-based model.

To investigate the different SHV alleles, the bla_SHV-1 gene derived from a clinical K. pneumoniae isolate (RKI346-12) was ligated into the pBT plasmid and cloned into a BL-free E. coli genetic background, strain XL1-Blue, to produce strain MK1. The curves of E. coli MK1 and the original clinical isolate, K. pneumoniae RKI346-12, as well as the expression levels of the bla_SHV-1 genes in both, were determined (Fig. 2) in order to characterize the model.

FIG 2.

Characteristics of the expression model. (A) Comparison of the growth curves of MK1 and RKI346-12 over 20 h. (B) Quantification of SHV-1 transcripts in the wild-type K. pneumoniae strain RKI346-12 and the E. coli MK1 mutant at two different growth phases. Significant differences were determined by two-way analysis of variance and the Bonferroni posttest, and significance was assumed when the P value was <0.05. **, P < 0.001; ***, P < 0.0001. Fold changes are given above the bars. (C). Isoboles of β-lactams combined with BLIs determined for MK1. The curves represent the combined concentrations of the first nonturbid well in a checkerboard assay.

The growth of the mutant MK1 was slightly delayed relative to that of RKI346-12, but both strains were in the transition phase after 6 h and in the stationary phase after 12 h (Fig. 2A), suggesting that growth differences in the gene expression model were not significantly impacted by differences in the growth stage between the model and the wild-type strain. The expression of bla_SHV-1 was 15.6-fold higher in MK1 than in RKI346-12 during the transition phase and 5.9-fold higher during the stationary phase (Fig. 2B), indicating that generally higher MIC values of the different substrates should be expected in this model. Thus, susceptibility and resistance could not be directly interpreted according to the European Committee on Antimicrobial Susceptibility Testing (EUCAST) breakpoints. Therefore, for a clone library based on MK1, the ratios of the MICs for the clones to the MICs for MK1 were determined.

The background strain, E. coli XL1-Blue, showed no resistance against any of the β-lactams tested. As expected, strain MK1 showed high MIC values for all penicillin derivatives, including amoxicillin (AMX) (MIC_AMX, >1,024 mg/liter), AMP (MIC_AMP, >512 mg/liter), and piperacillin (PIP) (MIC_PIP, 256 mg/liter), as well as for the first-generation cephalosporin CFZ (MIC_CFZ, 128 mg/liter), indicating SHV-1 resistance to these substrates. The MIC of the second-generation cephalosporin cefoxitin (FOX) was 8 mg/liter in both XL1-Blue and MK1, indicating SHV sensitivity to FOX. The MICs of the third-generation cephalosporins were almost similarly low for MK1 (MICs, ≤0.5 mg/liter) and XL1-Blue, except for the MICs of CAZ (MIC_CAZ, 1 mg/liter) and cefoperazone (CFP) (MIC_CFP, 8 mg/liter), which were 4-fold and 32-fold higher for MK1 than for XL1-Blue, respectively.

Other higher-generation cephalosporins, such as cefepime (FEP) and ceftaroline (CPT), and the monobactam ATM, as well as the carbapenems (imipenem [IMP], ertapenem [ETP], and meropenem [MEM]), were very effective against MK1 (MICs, <0.1 mg/liter); however, the MICs of FEP, CPT, and ATM were higher for MK1 than for XL1-Blue.

Due to the increased number of bla_SHV-1 transcripts, the effective concentrations of the BLIs CLA (MIC_CLA), sulbactam (SUL) (MIC_SUL), and TAZ (MIC_TAZ) were analyzed by checkerboard experiments combining different concentrations (0 to 64 mg/liter) of the recommended β-lactams with the specific BLIs (Fig. 2C). Although the combination of CLA with AMX (fractional inhibitory concentration index [FICI], 0.45 ± 0.2) or AMP (FICI, 0.33 ± 0.2) was synergistic, high CLA concentrations were necessary to noticeably reduce the MICs of both β-lactams in this model. Additionally, the formation of small-colony variants (SCVs), representing a slow-growing but antibiotic-tolerant alteration of vegetative bacteria, was observed, which might complicate the determination of the mutant phenotypes in this model. SUL was ineffective up to a concentration of 64 mg/liter when combined with AMP, and a reduction in the MIC_CFP resulted from the activity of CFP. Only TAZ efficiently reduced the effective MIC_PIP at the EUCAST-recommended concentration of 2 mg/liter.

Substitution mutant library.

To evaluate the impact of the positions and substitutions identified on the phenotype, a clone library was prepared by systematic site-directed mutagenesis of the pMK3 plasmid based on the 18 positions predicted by optiSLang and an additional 5 positions occurring in natural 2be or 2br variants that were not predicted to be relevant. The amino acid residues were exchanged based on those occurring in natural SHV variants. In this manner, 23 positions were mutagenized, and 27 single-substitution mutants were generated (unfortunately, we failed to generate mutants bearing the L169R and D179G substitutions).

We further combined various substitutions and generated eight double mutants (including one promoter mutation) and two triple mutants (Table S4). Among these mutations, the A25S mutation was combined with the M129V substitution (which was not predicted), since both substitutions occur in the 2be SHV-42 variant. Because the M129V substitution occurs in combination with T18A and/or L35Q in other SHV variants, including two 2b variants and unknown phenotypes, respectively, these substitutions were combined. The R43S substitution occurs solely combined with G238S in SHV-141, which is classified as a 2b variant in NCBI. Because G238S showed a high CoP value in relation to the 2be phenotype and has been reported as crucial for that phenotype (9, 10, 12), this double mutant was investigated in order to determine whether R43S reverts the variant to the 2b phenotype. The A187T substitution, predicted as 2br relevant because it seems to be wholly responsible for the 2br phenotype of SHV-26, also occurs in five variants with unknown phenotypes and in two 2b variants bearing the L35Q substitution. Thus, L35Q was combined with A187T. The A146V substitution is the only one in the 2be variant SHV-38 and was therefore predicted to be 2be relevant. This substitution also occurs in two 2b (SHV-71 and SHV-73) and two unknown SHV (SHV-144 and SHV-168) variants and in one 2br variant (SHV-72) accompanied by other substitutions, such as I8F or K234R, which were also combined with A146V in this study. The A146V substitution was also combined with the high-expression IS26 promoter.

The MICs of different penicillins (AMX, AMP, and PIP), BLIs (CLA, SUL, and TAZ), and first- and second-generation (CFZ, FOX), third-generation (CAZ, CTX, CRO, CFP, and cefpodoxime [CPD]), and higher-generation (FEP and CPT) cephalosporins, as well as those of the monobactam ATM and carbapenems (IMP, ETP, and MEM), were tested for the mutants and compared to those for the MK1 strain. An increase or reduction of >2-fold in the MIC was assumed to be significant due to the acceptable ±1 log₂ variation of the AST method, as recommended by EUCAST standard ISO 20776-1 (2006). The 2be phenotype of the mutants was assessed by the MIC changes of CAZ and CTX. Because CLA and SUL were insufficient in this model, we decided to use PIP-TAZ to interpret the 2br phenotype of the mutants.

Susceptibility to penicillins, first- and second-generation cephalosporins, and aztreonam.

The MICs of the penicillins for all the mutants were very similar to those for the MK1 strain, with two exceptions: the strains carrying the G238S (4-fold-reduced MIC_PIP) or D179N (6-fold-reduced MIC_PIP) substitution (Fig. 3A; Table S4). No significant changes in the MIC of the second-generation cephalosporin FOX were observed in the mutants, whereas the MIC_CFZ was reduced in the majority of the mutants (most dramatically in the T235A mutant, from 128 mg/liter in MK1 to 4 mg/liter). For the G238S mutant, the MIC_CFZ and MIC_ATM were increased 4-fold over those for MK1 (Fig. 3A).

FIG 3.

Changes in susceptibilities to different β-lactam antibiotics between the MK1 strain (SHV-1) and the substitution mutants. (A) Changes in the MICs of penicillins, first- and second-generation cephalosporins, carbapenems, and aztreonam (a monobactam) for single-substitution mutants and multiple-substitution mutants bearing the A146V substitution. (B) Changes in the MICs of selected β-lactams combined with BLIs for single-substitution mutants and multiple-substitution mutants bearing the A146V substitution.

Susceptibility to carbapenems.

For the SHV-38 allele, which bears only the A146V substitution, increased activity toward IMP has been reported (13). However, neither the single mutant nor double mutants bearing A146V showed differences in their susceptibilities to the carbapenems tested (Fig. 3A). Because the MIC depends on promoter activity, we coupled the strong IS26 promoter, which occurs naturally in plasmid-encoded SHV variants, with the A146V substitution. For this mutant, the MICs of ETP (0.023 mg/liter) and IMP (0.19 mg/liter) increased slightly, by 1.5-fold, over those for the MK1 strain.

Substitutions with relevance for BLI resistance.

In contrast to MK1, which showed a high MIC_CLA, most of the substitution mutants showed increased susceptibility to CLA (Fig. 3B), resulting in strongly reduced MICs of AMX and AMP when these drugs were combined with CLA. Like MK1, most of the single-substitution mutants were not susceptible to SUL, resulting in unchanged MICs for AMP and CFP in the presence of SUL. In the T235A mutant, the reduced MIC_CFP in the presence of SUL was associated with the generally reduced MIC_CFP (Fig. 4A). Increased susceptibility to SUL was conspicuous for the D104G, D179N, and T235A mutants and, to a lesser extent, was observed for the R43S, A187T, G238S, G238A, A243G, and R292Q mutants. A high MIC_TAZ was observed only with the M69I, M69L, E240K, and E240R substitutions. The substitution at position 69 also increased the MIC_SUL and restored the MIC_CLA, whereas position 240 was related solely to TAZ resistance. The K234R substitution led to a significantly increased MIC_AVI. This substitution and T235A also restored the MIC_CLA. Combining the A146V substitution with the strong IS26 promoter resulted in increased resistance to PIP-TAZ and CFP-SUL, but combining A146V with K234R increased susceptibility to CFP-SUL. No MIC changes for combinations with BLIs were observed for other multiple-substitution mutants (Table S4).

FIG 4.

Changes in the susceptibilities of the substitution mutants to selected cephalosporins. Shown are changes in the MICs for single-substitution (A) and multiple-substitution (B) mutants.

ESBL-relevant substitutions.

Relative to SHV-1, which showed a MIC_CAZ of 1 mg/liter, the majority of the substitutions led to at least a 4-fold reduction of the MIC_CAZ (Fig. 4A); only one substitution mutant, the D179N mutant, showed a 4-fold increase in the MIC_CAZ. This mutant also showed increased CTX, CRO, and FEP MICs. Significant increases in the MICs of most third-generation cephalosporins, except for CFP and CAZ, and also FEP and CPT, were observed with the G238S and G238A substitutions, whereas the E240K and E240R substitutions increased only the MIC_CPT. The MIC_CPT was also increased as much as 8-fold in several other mutants (the A25T and A25S, L35Q, A146T and A146V, L148V, and G156D mutants) but was strongly reduced in the M69I and T235A mutants. The MIC_CFP was slightly elevated in the M69L (but not M69I), D104G, A146T (but not A146V), and A243G mutants but was strongly reduced in the D179N and T235A mutants. The MIC_FEP was also reduced in the T235A mutant. Two-fold-increased activity against CTX was observed for the M129V, A243G, S270I, and R292Q mutants.

The substitution combinations produced did not result in any significant changes in the MICs of most of the cephalosporins tested or in the MICs of the monobactam ATM (Fig. 4B). The exceptions were the T18A M129V and T18A L35Q M129 mutants, for which the MIC_CFP was elevated relative to that for the M129V mutant, and the R43S G238S mutant, for which the MIC_CTX was halved and the MIC_CPD and MIC_CPT decreased by more than 42- and 512-fold, respectively, from those for the G238S mutant. The IS26 A146V promoter mutant showed strongly increased CFP and CPT MICs, a slightly increased MIC_FEP, and a restored MIC_CAZ.

DISCUSSION

We compared the MICs for MK1 with those in other reported SHV expression models to prove the suitability of the MK1 model for phenotype analyses. The MICs of selected third-generation cephalosporins, e.g., MIC_CAZ, for MK1 (1 mg/liter) ranged between the values observed for a strain carrying the high-copy-number pTZ18R plasmid used by Nüesch-Inderbinen et al. (1.5 mg/liter) (12) and those observed for a strain carrying the low-copy-number pCCR9 plasmid used by Randegger et al. (0.094 mg/liter for the weak promoter or 0.38 mg/liter for the strong promoter) (10). In contrast to our model, the model of Hujer et al., in which SHV-1 was subcloned into the pBC SK(–) plasmid and expressed in E. coli DH10B, showed higher resistance to TAZ than to CLA (9), which was also in agreement with the study of Payne et al., in which the SHV-1 enzyme activities of crude cell extracts were determined (14). A possible explanation might be altered expression of efflux pumps and porins of E. coli XL1-Blue, which might reduce the intracellular concentration of CLA, resulting in the resistance observed. However, this hypothesis was not proved. In addition, the formation of small-colony variants (SCVs) was observed within the inhibition zone of the CLA-AMX Etest strips. SCVs represent a dormant, slow-growing alteration of vegetative bacteria under stress, exhibiting an intrinsic tolerance to antimicrobials (15). Therefore, the 2br phenotype was assessed solely on the basis of PIP-TAZ MIC values. Although CLA was ineffective in this model, we classified the pBT-based model as generally suitable for our study.

In natural SHV BL variants, 114 positions are subject to permutation (not considering insertions), but only a few positions seem to be decisive for the phenotype. In other studies (10, 12), 14 positions were assigned to a specific phenotype or function of the SHV BLs: positions 146, 156, 164, 169, 179, 205, and 238 were assigned as crucial for resistance to third-generation cephalosporins (ESBL, or 2be, phenotype); positions 69, 130, and 244 were relevant for resistance to BLIs (2br); positions 104 and 240 were synergistic for the ESBL phenotype; and positions 166 and 235 were conserved (9, 10, 12, 16). By analyzing the naturally occurring NCBI-validated variants using mathematical models, we identified only 7 of these positions (positions 69, 104, 169, 179, 235, 238, and 240) and an additional 11 positions as related to the specific phenotypes (Table 1). These and five further sites (including positions 146 and 156) were mutagenized, and their phenotypes were characterized. Because positions 130, 166, and 244 were conserved in the validated SHV variants, they were classified as irrelevant. The R205L substitution was present only in the ESBL SHV-3 and was accompanied by the 2be-relevant G238S mutation, while the R164S substitution was present only in SHV-143, which is classified as a narrow-spectrum BL; therefore, these substitutions also were classified as irrelevant.

In accordance with previous studies (9, 10, 12), position G238 was predicted and proved to be the most crucial for the ESBL phenotype (2be). It has been postulated that replacement of the glycine at position 238 with larger amino acid residues, such as serine, results in rearrangements of the adjacent amino acid residues and enlargement of the active site, allowing bonding and hydrolysis of higher-generation cephalosporins (17). However, the G238S and G238A substitutions resulted in increased MICs of most third-generation cephalosporins tested (CTX, CRO, and cefpodoxime [CPD]), as well as of the fourth- and fifth-generation cephalosporins FEP and CPT. Thus, the small nonpolar, neutral alanine residue was also sufficient, even with CTX, CRO, and CPD MIC values notably lower than those with G238S, to enhance the catalytic activity toward many cephalosporins (except CAZ and CFP).

Among the 147 allelic SHV variants, 34 (23.13%) contain the G238S (n = 30) or G238A (n = 4) substitution; 21 of the G238S variants and 3 of the G238A variants are known ESBLs; 10 of these variants were not characterized yet, and 1 (SHV-141) was categorized as 2b. The substitutions at position 238 are accompanied by the E240K (n = 19) or E240R (n = 1) substitution in 58.8% of these variants, including SHV-141 and four of the variants of unknown phenotype (SHV-9, -154, -160, and -165). In our study, both the E240K and E240R substitutions increased only the MIC of CPT, but the E240K substitution has been reported to synergistically increase the MICs of CTX (9) and CAZ in natural variants when combined with the G238S substitution (10). Thus, it seems that SHV-141 has been misassigned and that all the unknown variants bearing both substitutions are most likely ESBLs.

To a lesser extent, and with different substrate specificity (CAZ, CTX, CRO, and FEP), position 179 could be confirmed as relevant to the 2be phenotype. Position 179 is the last residue of the Ω-loop, forming a salt bridge to its first position at 164. Disruption of this salt bridge is supposed to result in increased flexibility of the Ω-loop, allowing the bulky cephalosporins to be accommodated in the active site (18). In our model, as in the study by Hujer et al. (9), the D179N substitution was the only one leading to an increased MIC_CAZ. This single substitution occurs only in the SHV-8 variant, which has been shown to be potent against CAZ (10). Another single substitution, D179G, is present in SHV-24 and is classified as 2be. We failed to produce this substitution in our model, but it has been shown for SHV-24 that despite efficient binding of the substrate (low K_m), it hydrolyzes CAZ with a low catalytic rate (19). Thus, it seems that replacement of the acidic polar residue aspartic acid at position 179 with a neutral asparagine residue, but not with glycine, is crucial for CAZ resistance. The increase in the MIC_CAZ observed for the natural isolates bearing SHV-24 might be related to other factors, such as the porin alterations that are often observed in clinical isolates (20).

The M129V, A243G, S270I, and R292Q substitutions showed slightly increased activity against CTX (a 2-fold increase over the activity of SHV-1). Substitutions at these positions are found independently of other 2be-relevant substitutions in several 2be variants (M129V in SHV-42, A234G in SHV-40, S270I in SHV-98, and R292Q in SHV-148), but also in variants with 2b or unknown phenotypes. We hypothesize, on the basis of the elevated CTX activity, that these substitutions, combined with increased expression or other mechanisms, such as porin loss or increased efflux, might manifest as a 2be phenotype.

Several substitutions that are present in natural 2be SHV variants, but also in uncharacterized variants, at positions 25 (in 2 variants), 35 (in 61 variants), 146 (in 7 variants), 148 (in 1 variant), and 156 (in 7 variants) resulted in an increased MIC_CPT. Interestingly, CPT susceptibility was restored by combining A25S, L35Q, or A146V with other substitutions that were not related to an elevated MIC_CPT. A combination of the two substitutions A146V and G57D, which individually increased the MIC_CPT, resulted in susceptibility to CPT, indicating that some combinations of the substituted residues might suppress the CPT-resistant phenotype. In this study, we did not examine this phenomenon further, but the strong increase in the activity against CPT and the elevated activity against FEP in the A146T mutant when this substitution was combined with a strong promoter indicated a potential cross-resistance of SHV variants against higher-generation cephalosporins that is independent of the ESBL phenotype. SHV BLs are widespread in Gram-negative bacteria, particularly in K. pneumoniae. FEP and CPT are used for empirical therapy for some infections (e.g., urinary tract infections and hospital- and community-acquired pneumonia) in which the presence of SHV-expressing Gram-negative pathogens might represent an underestimated problem, because these phenotypes are not covered by most routine diagnostics for Gram-negative pathogens. This phenomenon must be evaluated in natural variants that usually contain other substitutions, which might influence the MICs of FEP and CPT. Particularly, it remains unclear how CPT, which exhibits extended R1 and R2 side chains relative to those of third-generation cephalosporins, becomes hydrolyzed by the SHV mutants, which did not show elevated MICs of other cephalosporins.

Increased activity against CFP resulted from the single substitutions M69L (in SHV-132 [unclassified]), D104G (in SHV-99 [2be]), A146T (in SHV-80 accompanied by L35Q [2b]), and A243G (in SHV-35 accompanied by L35Q and E89K [unclassified]; in SHV-40 accompanied by L35Q [2be]), none of which were associated with the 2be phenotype in our study. We hypothesized that individual substitutions that are not directly relevant might influence substrate specificity when combined with other substitutions or with a strong promoter, as was exemplified by the A146V substitution, which resulted in increased CFP and CPT MICs only when combined with the IS26 promoter. CFP is not and never was a guideline-recommended first-line cephalosporin for empirical treatment of infections with Gram-negative pathogens and thus did not reach a strong selective pressure, as can be expected for other cephalosporins, such as CAZ or CTX. These findings clearly show that various substitutions within the SHV variants that are currently considered neutral have the potential to mediate resistance to certain substrates, including future β-lactam derivatives.

Amino acids at two positions, M69 and E240, were strongly related to the BLI-resistant (2br) phenotype. The M69I substitution occurs in one characterized 2br allele (SHV-49) and in two uncharacterized variants (SHV-52 and -92), while M69L is present in the uncharacterized enzyme SHV-132. Both substitutions at position 69 led to resistance against TAZ, increased resistance to SUL, and restored resistance to CLA. In a study by Hujer et al., the M69I and M69L substitutions resulted in increased resistance to AMP-CLA (9), while Winkler et al. showed increased resistance to AVI as well for these substitutions (21). The discrepancies between our results and those of the other studies might be due to different models or different codon usage (we preferred the naturally occurring triplets with E. coli codon usage for generating the substitutions; in other works, this methodological detail was not specified).

Interestingly, both substitutions (E240K and E240R) at position 240, which was predicted to be 2be related, also strongly elevated resistance to TAZ, but only E240R restored CLA resistance. Neither of these mutations was found in any of the known 2br variants, and thus, this position has never been associated with the 2br phenotype. Substitutions at this position are present in 25 variants, of which only 8 are uncharacterized, while the others are ESBLs. Among the uncharacterized variants, four (SHV-9, 154, -160, and -165) contain a substitution at position 238, suggesting a 2be phenotype. Only one of remaining ESBLs contains only an E240K substitution accompanied by a L35Q mutation (SHV-31). Thus, we cannot exclude the possibility that position 240 plays a subordinate role in BLI resistance in natural variants. However, in the 2br variant SHV-26, only the A187T substitution is present, and this variant did not show any significant increase in the MICs of the BL-BLI combinations in our study. Thus, for both the natural variants SHV-26 and SHV-31, the reported phenotypes might be incorrectly assigned or related to other, additional mechanisms (such as porin loss, efflux, or promoter activity).

The S130G substitution found in SHV-10, which was shown to confer BLI resistance in other studies (9, 21, 22), was not investigated in the present study, because the SHV-10 variant has been removed from the recently updated BL database at NCBI; thus, this mutation was not considered by the algorithms.

The K234R substitution seemed to be associated with reduced susceptibility to AVI. Winkler et al. reported that this mutation also increased the MIC_AMP when AMP was combined with AVI or CLA (21, 22) but reduced the MIC_AMP when AMP was combined with SUL, proposing an acylation mechanism for this mutant; due to its close proximity to the catalytic serine 130, K234R alters the torsion angle of S130, disturbing the interaction with AVI or CLA (21). The K234R substitution is present in three SHV variants, of which two (SHV-72 and SHV-56) exhibit a 2br and one (SHV-73) exhibits a 2b phenotype. SHV-72 bears the additional mutations A146V and I8F. I8F in the SP has been suggested to increase the MICs of cephalosporins (10) due to enhanced secretion. The I8F A146V, A146V K234R, and I8F A146V K234R mutants showed resistograms very similar to that of A146V; thus, we cannot confirm the impact of residue F8 on the phenotype (Fig. 3B). In contrast, when A146V was combined with a strong promoter (IS26), the P_IS26-A146V mutant showed increased resistance to almost all β-lactam–BLI combinations tested, indicating that promoter activity plays a key role in the BLI-resistant phenotype of the SHV alleles. SHV-73, which is now assigned as 2b (NCBI database), has been reported formerly as 2br (www.lahey.org/Studies). Based on our data, we can exclude the frequent substitution L35Q (41.5% of all SHV alleles) that accompanies the K234R substitution in SHV-56 as 2br relevant. Thus, we hypothesize that the K234R substitution most likely confers latent resistance to BLIs that can be increased by a secondary mechanism, such as increased promoter activity.

In the study by Poirel et al., a purified SHV-38 protein that bears the A146V substitution showed measurable activity against IMP (turnover constant [k_cat] = 0.01/s), which was, however, very low compared to that against other natural SHV-38 substrates (e.g., for CAZ, k_cat = 110/s) (13). Thus, it is doubtful whether such a low activity toward carbapenems in SHV variants will confer clinically relevant resistance against carbapenems.

It must be noted that 16 of the mutants produced corresponded to naturally occurring variants, but only 8 of those (Table S5) exhibited the same phenotype as that reported in the NCBI database. Most discrepancies were found within the 2be and 2br phenotypes. Therefore, some stored entries in the databases seem to be error prone, falsifying the modeling. These findings might be the reason why the prediction accuracy of optiSLang modeling for the phenotypes was below 60%, indicating, in general, difficulties for the use of mathematical models. On the other hand, the mathematical models applied evidently found some latent positions that, in combination with other mechanisms, might confer resistance to different cephalosporins and BLIs. In many cases, the phenotype also depends on more than one mutation, and the combination of these could play a crucial role. This finding demonstrates the potential of mathematical models in finding new targets for molecular diagnostics and their predictive value for sequence-based analytics. However, the results also show clearly that mathematical models based only on phenotype and amino acid sequences tend to overstate neutral mutations. This is due mainly to the fact that in natural variants, but also in models, other intrinsic factors, such as promoter activity, different codon usages of the species, or additional resistance mechanisms (efflux/influx), play an important role in phenotype expression. This is difficult to capture in the common mathematical models. In the future, models based on whole-genome and transcriptome data using artificial intelligence could make significantly better predictions, so that the phenotypes could be derived directly from the genome data.

In conclusion, based on the collection of data for SHV BLs from the NCBI, 20 amino acid positions were identified in silico as phenotypically relevant; in vitro analysis revealed that only 2 positions each were crucial for the 2be (positions 179 and 238) and 2br (positions 69 and 240) phenotypes, but position 240 has not been found to be mutated in natural 2br variants so far. Our results demonstrated that the permutation and adaptation potential of the SHVs is far from exhausted. SHV enzymes that often accompany other BLs are currently an underestimated BL group in molecular resistance tests and AST. In the face of the widespread distribution and emergence of infections with Gram-negative bacteria, the ability of some SHV alleles to confer resistance to BLIs is of great clinical relevance. The BLIs combined with penicillins (e.g., PIP-TAZ) or cephalosporins (e.g., CFP-SUL or CAZ-AVI) represent a pillar of empirical therapy for infections with Gram-negative pathogens. In the recently published MERINO trial, the 30-day mortality of patients suffering from bloodstream infections by CRO-resistant E. coli or K. pneumoniae was significantly higher under PIP-TAZ treatment (12.3%) than under meropenem (MEM) treatment (3.7%) (23), indicating that for both empirical therapy and molecular diagnostics, BLI resistance should be taken into account. Recently, a new non-β-lactam inhibitor of carbapenemases, vaborbactam, was approved as fixed combination therapy with MEM for the treatment of complicated urinary tract or intra-abdominal infections as well as nosocomial lung infections by Gram-negative bacteria. Since carbapenemases, such as KPC, are often accompanied by SHVs, existing or rapidly acquired cross-resistance of SHV to vaborbactam should definitely be investigated in order to ensure the efficacy of the new therapy.

MATERIALS AND METHODS

In silico sequence analysis.

The curated amino acid sequences of 147 sulfhydryl-variable (SHV) allelic variants (including the signal peptide [SP] sequences) were downloaded from the NCBI BioProject database in September 2017 (https://www.ncbi.nlm.nih.gov/pathogens/beta-lactamase-data-resources/). The Bush-Jacoby-Medeiros (BJM) classification was provided by NCBI for 73 variants (29 2b, 39 2be, and 5 2br variants); 74 variants were still uncharacterized at the beginning of our study. The complete data set is available as Data Set S1 in the supplemental material.

Sequence alignment was performed by applying a Clustal Omega (24) algorithm for multiple sequence alignments using CLC Main Workbench, version 7.0.2 (Qiagen, Hilden, Germany). Based on this multiple alignment, continuous numbering was used for the mathematical models (see Data Set S1). The commonly used Ambler scheme (7) of the amino acid positions was used for the evaluation of the results and the discussion to improve comparability and traceability to other publications. The unprocessed SHV-1 protein (including the SP) was set as the reference sequence.

The algorithms were trained to predict phenotype-relevant amino acid substitutions (PRASs) or patterns of PRASs by applying different mathematical methods as described below based on the 73 well-characterized variants. Variants with unknown phenotypes were not included in the learning processes.

optiSLang method.

optiSLang (Dynardo GmbH, Weimar, Germany) was used to predict phenotype-relevant amino acid residues in SHV variants. optiSLang is a software for multidisciplinary tasks in parametric sensitivity analysis, multidisciplinary optimization, and reliability and robustness evaluation, as well as robust design optimization using stochastic analysis. The idea of the Metamodel of Optimal Prognosis (MOP) proposed by Most and Will is to reduce the complexity and variables by automatically applying well-known metamodels and tools for variable reduction (25). MOP is thus based on the search for the optimal set of input variables and the most suitable approximation model. The result is an approximation model that contains the most important variables.

To analyze the relevance of individual amino acid exchanges, the nominal SHV data set was transformed into a cardinal data set by the implemented Python script. Amino acid residues were numbered according to the multiple alignment of all 147 SHV variants, so that even delegations contained a pro forma position. The analyses were performed both by automatic approximation and by targeted preselection of the kriging regression model. Additionally, the input parameters were modified by including the isoelectric points of the amino acid residues and by including or excluding residues missing due to deletions. When missing residues were included, the variables of those positions were set at zero (since there are no values for a missing residue); when they were excluded, the position was left empty and an incomplete data set was used. The validation parameters were also either set automatically by optiSLang or varied, so that the modeling was performed with n – 1 variants (randomly chosen) and the cross-validation after each optimization run was performed on the one variant that was left out. In total, 11 analyses with different settings were performed and compared to each other (Fig. 1). After each optimization round, a coefficient of prognosis (CoP) was calculated. The CoP is equivalent to the coefficient of determination (R²) but applies to polynomial regression and thus is defined as 1 – [(sum of squared prediction errors per run)/(sum of squared total errors)] (25, 26). The evolutionary Pareto front algorithm iteratively optimized the model in subsequent optimization analyses, removing parameters without impact on the model. The optimizer settings were as follows: parameters, amino acid characteristics; no specific start design; criteria, maximum CoP; initialization, initial population size variable; end of analysis, after 5 to 90 generations or after 10 generations without changes. To visualize the relevance of the amino acid residues, only those that reached a CoP value (equivalent to PRAS positions) are shown in Fig. 1. In Table 1, the CoP values are given as mean values with standard deviation. (For more details, see the work of Most and Will [25].)

Microorganisms and culture conditions.

The clinical K. pneumoniae isolate RKI346-12 bearing an SHV-1 allele was provided by the Robert Koch Institute (RKI; Wernigerode, Germany). The shv-1 gene was confirmed by Sanger sequencing. The host E. coli strain XL1-Blue was purchased from Agilent Technologies (Santa Cruz, CA, USA). All strains and their derivatives were stored at –80°C in 10% glycerin stocks in Mueller-Hinton (MH) broth and were cultivated in MH broth or on MH agar plates (Roth GmbH, Karlsruhe, Germany) at 37°C overnight.

Oligonucleotides and enzymes.

Primers (Table S1) were designed using CLC Main Workbench, version 7.0.2 (Qiagen). Primers of high-performance liquid chromatography (HPLC)-purified quality were ordered from Sigma-Aldrich (St. Louis, MO, USA).

All restriction enzymes, DreamTaq Green DNA polymerase, thermosensitive alkaline phosphatase (AP), and T4 DNA ligase were purchased from Thermo Fisher Scientific, Inc. (Waltham, MA, USA). Sensiscript reverse transcriptase was purchased from Qiagen, and Platinum Taq polymerase was purchased from Invitrogen (Carlsbad, CA, USA). All enzymatic reactions were performed according to the manufacturers’ protocols using the corresponding buffers and supplements provided in the kits.

Cloning of shv genes.

K. pneumoniae RKI346-12 was used to amplify the bla_SHV-1 gene (including its natural promoter) by colony PCR using primers SHV-1-XhoI-for and SHV-1-BamHI-rev and a DreamTaq Green DNA polymerase kit. PCR products were extracted from 1.5% agarose gels after electrophoretic separation by a NucleoSpin Gel & PCR Clean-up kit (Macherey Nagel GmbH & Co. KG, Düren, Germany). The bla_SHV-1 PCR product and the pBT cloning vector (Agilent Technologies) were digested by XhoI and BamHI, and the plasmid was additionally dephosphorylated by AP before both were ligated using T4 DNA ligase to generate the new plasmid pMK3, which was transformed into competent E. coli XL1-Blue cells, resulting in strain MK1. Transformants were selected on MH agar plates supplemented with 2 μg/ml tetracycline and 100 μg/ml AMP. The plasmid was isolated by a Qiagen Plasmid Mini kit according to the manufacturer’s protocol and was confirmed by Sanger sequencing using the sequencing primers Seq-For-SHV-1 and Seq-Rev-SHV-1 (Table S1).

Mutagenesis of the shv gene.

Single-amino-acid substitutions in the bla_SHV gene were generated with pMK3, derived from strain MK1, and corresponding primers containing the respective mutations (Table S1) using the QuikChange II site-directed mutagenesis kit (Agilent Technologies) according to the manufacturer’s protocol. After transformation into supercompetent E. coli XL1-Blue cells, the transformants were selected on MH agar plates supplemented with 2 μg/ml tetracycline and 100 μg/ml AMP. From several transformants, the plasmids bearing the bla_SHV alleles were isolated using a NucleoSpin Plasmid kit (Macherey Nagel) according to the manufacturer’s protocol, and the plasmid sequences were confirmed by Sanger sequencing using primers MM-Seq-For-SHV-1 and MM-Seq-Rev-SHV-1 (Table S1) at Microgen Inc. (Seoul, South Korea). All mutants were named according to the substitutions introduced.

Antimicrobial susceptibility testing.

Standard laboratory powders of the antimicrobial agents or Etest strips were used in this study according to the manufacturers’ recommendations (for details, see Table S2). The MICs of the antibiotics were determined by the broth microdilution technique in accordance with EUCAST standard ISO 20776-1 (2006) by use of cation-adjusted MH broth or by Etest.

To prove the effective concentrations of BLIs in this model, checkerboard assays were performed combining CLA with amoxicillin (AMX) or AMP, TAZ with PIP, SUL with AMP or CFP, and AVI with AMP. Briefly, 100 μl of a 10⁶-CFU/ml bacterial suspension in MH medium was transferred to each well of a microwell plate (Greiner Bio-One) and was mixed with an equal volume of antimicrobial solution (1:2 dilution of the bacteria). The microwell plates were incubated at 37°C for approximately 16 to 20 h, and the concentrations of both antimicrobials correlating to the first nonturbid well in each row and column along the turbidity-nonturbidity interface were plotted against each other in order to obtain the isoboles and to visualize the interactions of the antimicrobials. Each test was performed at least in duplicate and included a growth control without the addition of any antibiotic.

qPCR.

The clinical isolate bearing the SHV-1 gene used for cloning and the corresponding E. coli transformant MK1 were used to determine the difference in bla_SHV-1 transcripts. Sterile MH broth was inoculated with a loop of fresh colony material; the optical density at 600 nm (OD₆₀₀) was adjusted to 0.081 (in a 1-cm cuvette, equivalent to a 0.5 McFarland standard); and the solution was subsequently diluted 1:300 in 10 ml of MH broth to achieve approximately 10⁵ bacteria/ml. The suspensions were grown at 37°C under rotation (150 rpm) in an orbital shaker until an OD₆₀₀ of 0.5 was reached. The cells were harvested by centrifugation at 5,000 rpm in a Beckman Microfuge R centrifuge, and total RNA was prepared using an RNeasy kit (Qiagen) according to the manufacturer’s protocol. RNA quality (RNA integrity number [RIN], >7) was assessed by a 2100 Bioanalyzer (Agilent Technologies), and RNA quantity was assessed by an Infinite M200 Pro multimode microplate reader (Tecan, Männedorf, Switzerland). The RNA (50 ng) was reverse transcribed into cDNA using the MM-shv1-T-BamHI-rev primer and Sensiscript reverse transcriptase. For quantitative PCR (qPCR), 1 μl of the cDNA was transferred to the PCR mixture containing the MM-shv1-T-BamHI-rev and MM-shv1-P-XhoI-for primers, 1.5 U/μl Platinum Taq polymerase, 250 μM deoxynucleoside triphosphate (dNTP) mixture (Carl Roth GmbH), and 0.15× SYBR green (Invitrogen) in a 25-μl total volume. qPCR was performed on a Rotor-Gene cycler (Qiagen) for 40 cycles, using 95°C for denaturation, 55°C for annealing, and 72°C for synthesis. To calculate the absolute copy number of the transcripts, the purified bla_SHV-1-bearing pMK3 plasmid was quantified using an Infinite M200 Pro microplate reader and a NanoQuant plate (both from Tecan, Männedorf, Switzerland). A serial dilution was used for qPCR for calibration. The absolute molecule number was assessed by applying the following equation:

$$\frac{\frac{X\text{g}}{\text{μl DNA}}}{\text{plasmid length in base pairs}\times \text{660}}\text{× 6}{\text{.022 × 10}}^{\text{23}}\text{=}Y\text{molecules/μl}$$

Data availability.

All data are included in this article or in the supplemental material.